Solvation of Phenylglycine- and Leucine-Derived Chiral Stationary

Pirkle , W. H.; Welch , C. J.; Zych , A. J. J. Chromatogr. 1993, 648, 101 ..... Zhao , C.; Cann , N. M. J. Chromatogr., A 2006, 1131, 110. [Crossref],...
0 downloads 0 Views 4MB Size
13022

J. Phys. Chem. B 2008, 112, 13022–13037

Solvation of Phenylglycine- and Leucine-Derived Chiral Stationary Phases: Molecular Dynamics Simulation Study Sorin Nita and Natalie M. Cann* Department of Chemistry, Queen’s UniVersity, Kingston, Ontario, Canada K7L 3N6 ReceiVed: October 27, 2007; ReVised Manuscript ReceiVed: June 28, 2008

A theoretical study of the solvation of (R)-N-(3,5-dinitrobenzoyl)phenylglycine- and (R)-N-(3,5-dinitrobenzoyl)leucine-derived chiral stationary phases (CSPs) is presented. Semiflexible models of the chiral selectors are prepared from B3LYP/6-311G** calculations, and these are used in the molecular dynamics simulations of the corresponding interface. The chiral interface is examined for four solvents: 100% hexane, 90:10 hexane: 2-propanol, 80:20 hexane:2-propanol, and 100% 2-propanol. Despite the similarities between phenylglycine and leucine, the interfaces are distinct both in terms of the selector orientations at the surface and in the number of hydrogen bonds formed with 2-propanol. We also find that an increase in alcohol concentration alters the preferred orientations of the selectors. I. Introduction Chiral high performance liquid chromatography (HPLC) is widely used to separate mixtures of chiral molecules.1 The selector in chiral HPLC can be introduced through chiral additives in the mobile phase, but it is more common to use a chiral stationary phase (CSP) where the selective molecule is attached to the support material. The two CSPs of interest in this study are brush-type, where the selective molecules are attached to the support material via a tether. They are based on N-(3,5-dinitrobenzoyl)phenylglycine (Figure 1a) and N-(3,5dinitrobenzoyl)leucine (Figure 1b), and they are prepared by a reaction of the dinitrobenzoyl-modified amino acid with a γ-aminopropylsilanized silica.2-4 These CSPs were found to be good at separating a wide range of π-donor and π-dipolecontaining analytes: R-naphthamides of primary and secondary amines,2 alcohols, diols, sulfoxides, heterocyclic compounds, and binaphthols,5-7 among others. Because of their versatility, they were among the first commercially available HPLC CSPs, having been introduced by Regis Technologies Inc. in the early 1980s. We will refer to these CSPs as DNB-phenylglycine and DNB-leucine, to emphasize that their structure contains an amino acid and a dinitrobenzoyl group. DNB-phenylglycine and DNB-leucine have common selectivity for some classes of compounds.2 However, DNB-leucine has an enhanced selectivity for compounds like benzodiazepines, whereas DNB-phenylglycine is usually used for separating arylsubstituted sulfoxides and derivatives of β-naphthol, R-indanol, R-tetratol, and aryl-substituted hydantoins.8 The chiral separation mechanism for these two CSPs has been extensively investigated and debated in the literature:7,9-12 the dinitrophenyl ring can form face-to-face ring interactions with a π-donor group in the analyte; the dinitrobenzamide NH (N(23)-H(24) in Figure 1a and N(21)-H(22) in Figure 1b) can hydrogen bond with the analyte; the oxygen from the other amide carbonyl (O(14) in Figure 1a,b) can act as a hydrogen-bonding site. The access to these three interaction sites is governed by the steric bulk of the phenyl or isobutyl groups located on the C(15) atom of the selectors.9 By means of molecular dynamics (MD) simulations, * Author to whom correspondence should be addressed. Telephone: (613) 533-2651. Fax: (613) 533-6669. E-mail: [email protected].

Figure 1. Semiflexible representations of the DNB-phenylglycine (a) and DNB-leucine (b) selectors with the rigid units indicated by shaded areas. The silicon atoms numbered 1-3 form part of the underlying silicon layer representative of the surface. The atom numbering shown here is used throughout this article.

we examine the DNB-phenylglycine and DNB-leucine conformations at the interface and relate these to H-bonding abilities of the selectors. Simulations of interfaces present many challenges. Stationary phase interfaces are particularly demanding because they require that the selector have a complete, atomic level description, together with a proper representation of its flexibility, and with an appropriate representation of the surface environment (surface coverage, end caps, etc.).13 Nonetheless, some simulations of achiral selectors for HPLC have been published.14-17 Most notably, the entanglement and solvation of long-chain alkanes tethered to silica has been examined in some detail.14-17 To date, only a few simulations of chiral interfaces have been published: We have previously examined the solvation of N-(1-phenylethyl)-N′-[3-(triethoxysilyl)propyl]urea18,19 and the Whelk-O1 CSP.20,21 In this article, the interfaces include the selectors (DNB-phenylglycine or DNB-leucine), trimethylsilyl end caps, and silanol groups and these are covalently attached to a single underlying layer of Si that is stationary throughout the simulation. The selectors, end caps, and silanol groups have

10.1021/jp710392k CCC: $40.75  2008 American Chemical Society Published on Web 09/24/2008

Solvation of Chiral Stationary Phases surface coverages of 1.10, 3.29, and 4.26 µmol/m2, respectively, consistent with experimental coverage for this type of CSP.2,22,23 In normal-phase chiral chromatography, the mobile phase is usually a mixture of hexane and 2-propanol. The amount of alcohol is carefully optimized since it is well known that solvent composition has an effect on the absolute retention times because of competitive hydrogen bonding with the substrate:24 Typically the retention time decreases as the alcohol percentage in the mobile phase is increased.25 The DNB-phenylglycine and DNBleucine CSPs are usually used in conjunction with a hexane:2propanol solvent mixture, with the percentage of 2-propanol varied between 0 and 20% depending on the polarity of the analyte that is separated.7,24 To fully understand the impact of solvent composition, we examine the interfaces for four solvent compositions: 100% hexane, 90:10 hexane:2-propanol, 80:20 hexane:2-propanol, and 100% 2-propanol. The MD simulations reveal the solvent distribution in the vicinity of the DNBphenylglycine and DNB-leucine interfaces, and emphasis is placed on identifying and characterizing hydrogen-bonding interactions with the solvent. Other techniques for modeling chiral stationary phases have been summarized elsewhere,26-28 and only a brief overview is given here. Statistical approaches that provide structure-selectivity relationships have been applied to chiral chromatography.29-31 These methods have the important advantage that they provide a quick prediction of separation efficiency. However, their range of applicability is restricted to analytes that are closely related to those upon which the structure-selectivity relationship was originally designed. A docking study typically examines single host-guest complexes that are either isolated or embedded within a dielectric continuum. Some time ago, Lipkowitz et al.32-35 completed a number of molecular mechanics and energy minimization studies in order to determine conformations of DNB-phenylglycine and predict the mechanism by which the chiral separation occurs. From MM2 force field calculations, the CSP was found to have two important conformations that are separated by only 5 kcal/mol.33 Using the lowest-energy structures for DNB-phenylglycine and 1-(9-anthryl)-2,2,2trifluoroethanol, Lipkowitz33,34 predicted the correct retention order found in chromatographic experiments and derived a binding mechanism that agreed with the proposed three point model.9,10 Similar studies have been performed for DNBleucine.36 Together, these studies provide an important first step in understanding these CSPs but the complexity of the analyte-CSP interaction meant that retention order was not always predicted successfully. As noted in a review article,32 “the best approach is to carry out full molecular simulations rather than simple energy minimizations.” We begin here with a detailed simulation-based examination of the chiral interfaces, including solvent. The force fields for DNB-leucine and DNBphenylglycine have been derived based on an extensive series of ab initio calculations at the B3LYP/6-311G** level of theory. Major conformers are identified and torsions, bends, improper torsions, and bond stretches are examined to ascertain the range of motion that can be expected. Finally, semiflexible models suitable for molecular dynamics simulations are obtained. This article begins in section II with a description of the DNBphenylglycine and DNB-leucine models derived from ab initio calculations. Details of the molecular dynamics simulations are also provided. The interfacial structure, solvent distributions, and hydrogen bonding at the interface are considered in section III. The article concludes with a brief discussion of the results.

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13023 II. Theoretical Details II.1. Representation of the Chiral Selectors. The DNBphenylglycine and DNB-leucine selectors are shown in Figure 1. Throughout this article, the term “selector” will be used to refer to a single chiral molecule of a CSP. The two selectors consist of a dinitrophenyl (DNP) group, a chiral carbon located between two amide linkages, a lateral group (a phenyl ring for DNB-phenylglycine and an isobutyl chain for DNB-leucine), and a tether that joins the selector to the surface. As shown in Figure 1, the model tether is trifunctionally bonded to the surface. Experimental surfaces may also include mono- and bifunctionally linked selectors,5 but these are undesirable since they are less resistant to hydrolysis. There are multiple hydrogenbonding sites within each selector: the oxygen and hydrogen atoms in the amide groups are of particular interest but the nitro groups are also examined. Figure 1 shows the atom numbering which will be used throughout this article. Semirigid representations of both CSPs are employed in our simulations. To be precise, the dinitrophenyl rings in both selectors and the phenyl ring in DNB-phenylglycine are rigid. All other portions of the selectors are flexible. These rigid regions are referred to as rigid units (RUs), and in our simulations, their translation and rotation is consistent with the forces on the atoms within the unit. However, the relatiVe atomic positions within the unit do not vary with time. In defining an interaction potential, we omit intramolecular potentials within the RU since these will be constant. It is important to note that the RU atoms still interact with atoms outside the unit, including atoms within the same selector, from other selectors, from the end caps, and so on. RUs are also involved in intramolecular potentials that include non-RU atoms. For example, in the DNBphenylglycine selector the potential includes a bend for C(15)-C(17)-C(18) and a torsion for C(13)-C(15)-C(17)C(18). The shaded regions in Figure 1 identify the RUs for DNB-phenylglycine and DNB-leucine. The energy of the interfacial system is given by

Efull )

1 2

∑ mibV i2 + 21 ∑ mRUbV RU2 + 21 ∑ IRUωbRU2 + i

RU

RU

∑ ULJ + ∑ Uelectro + ∑ Uintra (1) The first term is the translational kinetic energy of all the atoms, except those involved in RUs. The second and third terms are the translational and rotational kinetic energies, respectively, of the rigid units. mRU, IRU, and ω b RU are the total mass, moment of inertia, and angular velocity of the rigid unit, respectively. The last three terms in eq 1 are the Lennard-Jones potential (ULJ), the electrostatic potential (Uelectro), and the intramolecular potential (Uintra). The Lennard-Jones (LJ) potential between atoms i and j is

[( ) ( ) ]

ULJ ) 4εij

σij rij

12

-

σij rij

6

(2)

where rij, εij, and σij are the interatomic separation, the LJ well depth, and the LJ length parameter, respectively. Ewald sums are employed to evaluate Uelectro, as discussed in section II.3. The intramolecular potential, Uintra, is given by

Uintra ) Ustretch + Ubend + Utorsion + Uimproper

(3)

where energetic contributions correspond to bond stretching, angle bending, torsions, and improper torsions. The bond stretching potential is

13024 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Ustretch ) kst(r - re)2

Nita and Cann

(4)

where kst is the stretching force constant, r is the distance between the two atoms, and re is the equilibrium bond length. Within the simulations bonds are kept fixed using the Rattle algorithm,37 except for bonds joining RUs to the flexible portions of the molecule. The bending potential is given by

Ubend ) kθ(θ - θe)2

(5)

where kθ, θ, and θe are the force constant, the angle, and the equilibrium angle, respectively. Bends are considered between all covalently linked sets of three atoms, except when all three atoms belong to an RU. Improper torsion potentials have the form

Uimproper ) kω(ω - ωe)2

(6)

where ω is the out-of-plane angle and ωe is the corresponding equilibrium value. The out-of-plane angle for an atom D surrounded by three atoms A, B, and C is defined as the angle between the vectors normal to the two planes DAB and ABC. Improper torsion is used to restrict out-of-plane bending motions at amide nitrogens, amide carbons, and the ring carbons joining an RU to the rest of the molecule. A modified Ryckaert-Bellemans38 potential is adopted for torsions: torsion

U

)

Ntorsions

6

j)1

i)0

∑ ∑ cij cosi(φj + φij)

(7)

where, for the jth torsion, cij is the coefficient of the ith term, φj is the dihedral angle, and φij is the corresponding phase shift. Torsions are considered between all covalently linked sets of four atoms, except when three or more atoms belong to an RU. DNB-phenylglycine and DNB-leucine differ only in the identity of the lateral group. Based on this, one might expect that the potential parameters in eqs 4-7 would be very similar for both. In particular, one might expect identical, or near identical, backbone parameters. However, the differences in their experimentally observed selectivities5-7 suggests otherwise. According to the three point binding model, the lateral groups do not directly participate in the selectivity,34,36,39 and thus the conformations of the two selectors at the interface differ such that selector-analyte interactions are impacted. Experimental studies of related compounds also indicate important conformational differences. Chin et al.40 summarized the results of infrared (IR) spectroscopy studies on the impact of lateral groups based on gas phase isolated dipeptides. These dipeptides share a common backbone with the selectors of interest here. Interestingly, Chin observed a change in the optimal backbone structure when the lateral group is aromatic. Gung et al.41 also noted, based on the results of NMR and IR studies, that bulkier alkyl substituents lead to a change in conformational probabilities, such that new backbone configurations appear. With these considerations in mind, we have proceeded by carefully evaluating the intramolecular parameters for both selectors rather than adopting published parameters. Simulations of the full chiral interfaces requires appropriate definitions of the potential terms for the selectors, end caps, silanol groups, solvents, and underlying Si layer. The DNB-phenylglycine and DNB-leucine selectors are considered in sections II.1.a and II.1.b, respectively, and all other components are discussed in sections II.2 and II.3.

II.1.a. The DNB-phenylglycine Model. The major conformers of DNB-phenylglycine were obtained42 from geometry optimizations at the MP2/6-311G**, B3LYP/6-311G**, and HF/ 6-311G** levels of theory. All ab initio calculations are based on the individual, isolated selector, and since the tether is not expected to directly impact upon CSP solvation and chiral discrimination, the propylsilyl group is replaced with a methyl for the ab initio calculations. A similar truncation has been adopted by others.34,36,43 All three ab initio methods identify the same global minimum and two local minima within 20 kJ/ mol of the global minimum. Additional minima, at higher energies, were also identified, but these are less relevant for the simulations and will not be discussed further. It is worth noting that MM2 force field optimizations43 identified similar structures for the global and lowest-energy local minimum, although the relative energy was considerably smaller than the ab initio values. The three lowest-energy structures, obtained from the B3LYP calculation, are shown in Figure 2, along with the energies relative to the minimum. The MP2 and B3LYP methods lead to nearly identical structures, with all dihedral angles agreeing to within 10° and an average difference in the dihedrals of only 5°. The HF structures are also in close agreement except for the structure in Figure 2b, where HF predicts that the lower amide group is twisted relative to the B3LYP and MP2 structures. The global minimum (Figure 2a) is characterized by the C(13)-O(14) and N(23)-H(24) bonds pointing in the same direction and almost in the same plane (dihedral angle O(14)-C(13)-N(23)-H(24) is -2.9°). However, the O(14)H(24)-N(23) angle is only 111°, suggesting that there is poor intramolecular H-bonding. In the energetically nearest local minimum, shown in Figure 2b, the N(11)-H(12) and C(25)O(26) bonds appear close, but the N(11)-H(12)-O(26) angle is only 139°, thereby suggesting, at best, a strained hydrogen bond between these groups. Overall, the energetically nearest local minima result from backbone twists either above or below the chiral carbon. The DNB-phenylglycine interaction potential should reproduce, as best as possible, the energetically accessible conformers of the selector along with the energetic costs of conformational changes. With this in mind, Figure 3 outlines the iterative, eightstep process used to obtain the intramolecular potential for the truncated selector. Since the B3LYP and MP2 calculations are in good agreement for the conformer structures and relative energies, all further calculations employ the more efficient B3LYP functional with the 6-311G** basis set. Consider first the stretching (eq 4), bending (eq 5), and improper torsion (eq 6) potentials. In step 1, single point calculations around the equilibrium values are performed. The resulting energies are fitted to obtain the potential parameters in step 2. Specifically, least-squares fits are performed for bonds joining the RUs to the flexible portions of the selector. The resulting force constants are 1.355 × 105 kJ · mol-1 · nm-2 and 1.106 × 105 kJ · mol-1 · nm-2, for the C(15)-C(17) and C(25)-C(27) bonds, respectively. The relevant equilibrium bond lengths can be obtained from Table 1. The full set of bending potential parameters is given in Table 2. The equilibrium angles are obtained from the global energy minimum, and the corresponding force constants are obtained from a least-squares fit to energy calculations for angles within θe ( 5°, or from a known force field,44,45 as indicated in the table. Similarly, leastsquares fits yield the improper torsion parameters given in Table 4.

Solvation of Chiral Stationary Phases

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13025

Figure 2. Stable conformers of the DNB-phenylglycine and DNB-leucine selectors identified from B3LYP/6-311G** calculations. (a) DNBphenylglycine global minimum, (b, c) DNB-phenylglycine lowest energy local minima, (d) DNB-leucine global minimum, (e, f) DNB-leucine lowest energy local minima. Energy differences are reported relative to the appropriate global minimum from B3LYP/6-311G**, MP2/6-311G**, and HF/6-311G** calculations. Oxygen, nitrogen, carbon, and hydrogen atoms are shown in red, blue, gray, and white, respectively.

Steps 3 and 4 (Figure 3) are, respectively, the assignment of Lennard-Jones parameters (see eq 2) and the evaluation of atomic charges for the electrostatic potential. The OPLS force field46-48 is used for the LJ parameters of all atoms, except for silicon where OPLS parameters are not available and CHARMM44 values have been used instead. Lorentz-Berthelot rules49,50 are used for all pairs of distinct atoms. CHELPG51 atomic charges have been evaluated from the B3LYP/6-311G** global minimum. The resulting charges were compared to OPLS46-48 values, and overall, the differences are small. Of course, the CHELPG charges are correctly nonzero for the DNP carbons but transferable OPLS parameters cannot capture this. This is noteworthy since π-π interactions with the DNP ring are expected in the three point binding model9-12 of DNBphenylglycine. CHELPG atomic charges were also evaluated for the local minima and compared to the selected set. Again, the differences are small. The iterative portion of the model parametrization couples two factors. First, Lennard-Jones and electrostatic potentials are evaluated for all intermolecular and intramolecular atomic pairs, except for intramolecular atoms that are “close”. To be precise, these potentials are evaluated for all intramolecular pairs of atoms separated by four or more bonds, and pairs of atoms separated by only one or two bonds are considered “close”. For atoms separated by three bonds, so-called 1-4 atomic pairs, a scaling factor is introduced52,53 so that a fraction of these energetic contributions is included. For the selectors of interest here, 1-4 scaling is particularly critical due to the proximity of the amide groups. Specifically, the backbone conformation is sensitive to the strength of the interaction between the amides, and this interaction is directly impacted by the chosen scaling

factors. For instance, with the 1-4 scaling factors set to 0, the lowest-energy structures are similar to Figure 2a,e. On the other hand, when the scaling factor is closer to 1, the lowest-energy structures are similar to Figure 2b,d. Clearly, the 1-4 scaling parameters must be optimized52,53 to yield the correct balance of backbone structures. Second, torsional motion introduces significant changes in the molecular shape and selector-specific torsion potentials were derived as follows. For each dihedral, 11-18 restricted geometry optimizations were performed with the chosen angle given a fixed value between 0 and 360°, while the other parameters were fully optimized. This allows for structural relaxation, an essential step since the backbone consists of a series of single bonds, so that the torsions are strongly coupled. As a result, the energies obtained from restricted geometry optimizations are viewed as functions of all the dihedral angles and the parameters of eq 7 are obtained from a simultaneous least-squares fit54 to 94 energies. Prior to data analysis, the energies are corrected21,55 for Lennard-Jones, electrostatic, bend, and improper torsion contributions to yield corrected energies that reflect the torsional contributions to the total energy. In this way, the model will reproduce the correct relative energies when the various contributions are combined to yield the total potential energy (see eqs 1 and 3). Importantly, this correction depends on the 1-4 scaling factors chosen for the electrostatic and LJ contributions. An iterative cycle consists of selecting scaling factors for the 1-4 atomic pairs, correcting the torsional energies, simultaneous fitting of all torsional potentials, and model assessment. The latter is based on energy minimization following a combination of sequential and Monte Carlo conformational changes. Mini-

13026 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Nita and Cann

Figure 3. Detailed explanation of the parametrization procedure for the DNB-phenylglycine model.

mizations begin from a total of 33 000 randomly generated starting structures. Following this, the optimized structures are compared to the B3LYP/6-311G** global energy minimum and the lowest energy local minimum (Figure 2a,b). A new cycle begins with the examination of different 1-4 scaling factors. Since the ab initio model selector does not include the tether, literature potentials for hydrocarbon chains56-58 were employed for torsions within the tether. However, the C(8)-C(9)-C(10)N(11) torsion at the top of the tether is evaluated from B3LYP/ 6-311G** calculations of N-[3-(trihydroxysilyl)]propyl]formamide, a truncated analogue of the lower portion of the selector. The full set of torsion parameters is given in Table 3. Scaling factors 0.7 and 0.0 for the 1-4 Lennard-Jones and electrostatic potentials, respectively, yield the best agreement with ab initio results. Specifically, the optimal parameters yield the correct global energy minimum structure and a structure very similar to Figure 2b, roughly 9 kJ/mol higher in energy. In between these two structures, a few additional minima are found. First, one structure having the global energy minimum backbone with a small twist in the DNP group has been identified. Second, a

somewhat twisted backbone structure also appears as a minimum for the model. This structure is similar to the global energy minimum, and in ab initio calculations, this structure optimizes to the global. With the addition of the tether, and in the presence of the silanol and end caps at the surface, this minimum adopts surface conformations similar to those of the global. As a result, we have not considered it further. II.1.b. The DNB-leucine Model. The major conformers of DNB-leucine were obtained42 from geometry optimizations at the MP2/6-311G**, B3LYP/6-311G**, and HF/6-311G** levels of theory. All three methods identify the same global minimum and two local minima within 5 kJ/mol. This small energy difference is within the level of accuracy of all three methods59 and highlights the important difference between DNB-phenylglycine and DNB-leucine. Specifically, the flexible sec-butyl side chain in the latter leads to additional minima very close in energy to the global minimum. In fact, eight local minima were identified with energies within 20 kJ/mol of the global minimum, and still more are expected based on torsions of the sec-butyl group. Not surprisingly, Gallagher et al.60 identified two different

Solvation of Chiral Stationary Phases

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13027

TABLE 1: Details of the DNB-phenylglycine and DNB-leucine Modelsa DNB-phenylglycine

DNB-leucine

atom no.

atom/group

ε (kJ/mol)

σ (nm)

q (|e|)

atom/group

ε (kJ/mol)

σ (nm)

q (|e|)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38

Si Si Si O O O Si CH2 CH2 CH2 N H C O C H C CH CH CH CH CH N H C O C CH C CH C CH N O O N O O

2.510 2.510 2.510 0.636 0.636 0.636 2.510 0.494 0.494 0.494 0.711 0.193 0.439 0.879 0.276 0.126 0.293 0.460 0.460 0.460 0.460 0.460 0.711 0.193 0.439 0.879 0.293 0.460 0.293 0.460 0.293 0.460 0.502 0.711 0.711 0.502 0.711 0.711

0.392 0.392 0.392 0.315 0.315 0.315 0.392 0.391 0.391 0.391 0.325 0.040 0.375 0.296 0.350 0.250 0.355 0.375 0.375 0.375 0.375 0.375 0.325 0.040 0.375 0.296 0.355 0.375 0.355 0.375 0.355 0.375 0.325 0.296 0.296 0.325 0.296 0.296

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.25 -0.48 0.21 0.49 -0.48 0.17 0.04 0.23 -0.10 0.06 -0.04 0.08 -0.15 -0.54 0.27 0.57 -0.51 -0.02 0.06 0.02 0.02 0.05 0.01 0.66 -0.38 -0.38 0.64 -0.38 -0.38

Si Si Si O O O Si CH2 CH2 CH2 N H C O C H CH2 CH CH3 CH3 N H C O C CH C CH C CH N O O N O O

2.510 2.510 2.510 0.636 0.636 0.636 2.510 0.494 0.494 0.494 0.711 0.193 0.439 0.879 0.276 0.126 0.494 0.335 0.670 0.670 0.711 0.193 0.439 0.879 0.293 0.460 0.293 0.460 0.293 0.460 0.502 0.711 0.711 0.502 0.711 0.711

0.392 0.392 0.392 0.315 0.315 0.315 0.392 0.391 0.391 0.391 0.325 0.040 0.375 0.296 0.350 0.250 0.391 0.385 0.391 0.391 0.325 0.040 0.375 0.296 0.355 0.375 0.355 0.375 0.355 0.375 0.325 0.296 0.296 0.325 0.296 0.296

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.22 -0.52 0.31 0.53 -0.51 0.14 0.05 -0.01 0.23 -0.10 -0.08 -0.54 0.30 0.57 -0.51 -0.03 0.07 0.02 0.05 0.03 0.01 0.65 -0.38 -0.38 0.66 -0.38 -0.38

a Atomic charges are obtained using the CHELPG algorithm applied to the B3LYP/6-311G** global energy minimum. Lennard-Jones energy and length parameters are taken form the OPLS46-48 and CHARMM44 force fields.

backbone conformations present in crystals of DNB-leucine. Lipkowitz et al.39 also noted the multitude of minima in their force field (MM2D) study of a DNB-leucine analogue. The global minimum, and the two nearly isoenergetic local minima, are shown in Figure 2. The figure shows the B3LYP structures, but all three methods yield nearly identical structures: the average difference between the dihedrals predicted from B3LYP and MP2 is only 5.4°. Consider the global and local minima for DNB-leucine. The global energy structure for DNB-leucine (Figure 2d) closely resembles the local minimum of DNB-phenylglycine shown in Figure 2b. The converse is also true, and the DNB-leucine local minimum in Figure 2e has a backbone structure that is very close to the global energy structure of DNB-phenylglycine (Figure 2a). Several points are worth noting with respect to this exchange. First, B3LYP calculations predict that the two backbone structures differ by 10.86 kJ/mol for DNB-phenylglycine but only by 1.88 kJ/mol for DNB-leucine. Second, for DNB-leucine, torsions within the lateral group will lead to several more energetically accessible structures, for a given backbone structure, but this will not occur for DNB-phenylglycine. Finally, the simple replacement of a phenyl ring with an isobutyl group has altered the energy landscape such that a local minimum for DNB-phenylglycine becomes the global

minimum for DNB-leucine, and vice versa. IR studies of gas phase dipeptides offer an instructive analogue.40 When the residue is small (a methyl group for alanine, for example), the C7 conformer is favored, followed by a C5. The former is comparable to the backbone structures in Figure 2b,d while the latter is analogous to Figure 2a,e. In contrast, an aromatic residue interchanges this order such that C5 is favored. Thus, the interchange observed for the lowest-energy backbone conformations of DNB-leucine and DNB-phenylglycine is directly analogous to the experimental trend observed for dipeptides. The development of a classical potential for the intramolecular motion of DNB-leucine proceeds analogously to Figure 3. However, a few more bends and torsions are required to represent the motion of the lateral isobutyl group. In particular, torsional potentials were obtained from a leastsquares fit to energies obtained from 110 restricted geometry optimizations. To briefly summarize the results, for DNBleucine bond stretching is only considered between C(23) and C(25) and, based on a least-squares fit to ab initio calculations, the force constant is 1.123 × 105 kJ · mol-1 · nm-2 and the corresponding equilibrium bond length can be obtained from Table 1. The full set of bending, torsion, and improper torsion parameters is given in Tables 2-4. CHELPG51 atomic charges are given in Table 1.

13028 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Nita and Cann

TABLE 2: Details of the Bending Potentials for the DNB-phenylglycine and DNB-leucine Modelsa DNB-phenylglycine

a

angle

θe (rad)

7, 4, 1 7, 5, 2 7, 6, 3 8, 7, 4 8, 7, 5 8, 7, 6 9, 8, 7 10, 9, 8 11, 10, 9 12, 11, 10 13, 11, 10 13, 11, 12 14, 13, 11 15, 13, 11 15, 13, 14 16, 15, 13 23, 15, 16 17, 15, 16 23, 15, 13 17, 15, 23 17, 15, 13 18, 17, 15 22, 17, 15 24, 23, 15 25, 23, 24 25, 23, 15 26, 25, 23 27, 25, 26 27, 25, 23 28, 27, 25 32, 27, 25

2.20 2.20 2.13 1.96 1.97 1.84 2.01 1.96 1.98 2.07 2.14 2.06 2.17 2.00 2.11 1.89 1.89 1.89 1.86 1.97 1.96 2.10 2.10 1.98 2.12 2.10 2.15 2.10 2.03 2.04 2.16

DNB-leucine -1

kθ (kJ mol

-2

angle

θe (rad)

kθ (kJ mol-1 rad-2)

source

7, 4, 1 7, 5, 2 7, 6, 3 8, 7, 4 8, 7, 5 8, 7, 6 9, 8, 7 10, 9, 8 11, 10, 9 13, 11, 10 12, 11, 10 13, 11, 12 14, 13, 11 15, 13, 14 15, 13, 11 16, 15, 13 21, 15, 16 21, 15, 17 17, 15, 13 21, 15, 13 17, 15, 16 22, 21, 15 23, 21, 22 23, 21, 15 25, 23, 21 24, 23, 21 25, 23, 24 26, 25, 23 30, 25, 23 18, 17, 15 20, 18, 17 20, 18, 19 19, 18, 17

2.20 2.20 2.13 1.96 1.97 1.84 2.01 1.96 1.98 2.14 2.08 2.04 2.17 2.12 1.98 1.90 1.84 1.91 1.96 1.91 1.94 2.04 2.09 2.13 2.04 2.15 2.10 2.04 2.16 2.01 1.91 1.93 1.97

142.26 142.26 142.26 209.10 209.10 209.10 259.81 259.81 259.81 481.48 502.70 385.04 790.30 796.65 820.43 349.99 375.50 500.45 544.44 545.62 331.83 357.37 346.67 567.11 736.92 841.78 845.15 634.27 635.98 440.86 512.33 494.41 543.65

ref 44 ref 44 ref 44 ref 45 ref 45 ref 45 ref 45 ref 45 ref 45 this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work this work

rad )

142.26 142.26 142.26 209.10 209.10 209.10 259.81 259.81 259.81 349.31 484.32 347.98 884.04 762.56 884.20 295.79 354.74 352.90 720.08 566.03 532.70 545.96 545.89 330.89 324.13 453.23 792.59 792.53 735.36 639.69 638.52

The relevant angles are identified by three atoms, following the numbering in Figure 1.

TABLE 3: Details on the Torsional Potentials for DNB-phenylglycine and DNB-leucine used in the Simulationsa torsion

c0

c1

c2

c3

c4

8, 7, 4, 1 8, 7, 5, 2 8, 7, 6, 3 9, 8, 7, 4 10, 9, 8, 7 11, 10, 9, 8 13, 11, 10, 9 15, 13, 11, 10 23, 15, 13, 11 25, 23, 15, 13 27, 25, 23, 15 28, 27, 25, 23 18, 17, 15, 13

7.949 7.949 7.949 6.983 8.231 8.231 0.890 12.620 12.620 12.620 12.620 12.620 12.620

7.892 7.892 7.892 17.736 16.952 16.952 -0.195 7.803 14.634 -8.766 -59.262 2.500 8.011

2.722 2.722 2.722 0.886 1.133 1.133 -0.821 18.924 -42.180 -7.187 -35.165 -17.942 -2.880

DNB-phenylglycine -18.564 -18.564 -18.564 -25.606 -26.317 -26.317 7.510 10.689 -48.107 14.589 -70.642 47.520 81.300 -57.459 138.503 6.517 -6.181 104.854 -18.618 -2.976

8, 7, 4, 1 8, 7, 5, 2 8, 7, 6, 3 9, 8, 7, 4 10, 9, 8, 7 11, 10, 9, 8 13, 11, 10, 9 15, 13, 11, 10 21, 15, 13, 11 23, 21, 15, 13 25, 23, 21, 15 26, 25, 23, 21 18, 17, 15, 13 19, 18, 17, 15

7.949 7.949 7.949 6.983 8.231 8.231 0.890 5.803 5.803 5.803 5.803 5.803 5.803 5.803

7.892 7.892 7.892 17.736 16.952 16.952 -0.195 12.441 4.289 -19.302 -41.804 0.404 -26.127 18.580

2.722 2.722 2.722 0.886 1.133 1.133 -0.821 34.663 56.190 -11.580 -26.999 2.635 -27.795 33.246

DNB-leucine -18.564 -18.564 -18.564 -25.606 -26.317 -26.317 7.510 10.689 -44.216 -6.107 -21.161 -103.841 8.157 83.892 113.907 32.088 -2.336 -124.109 64.710 127.153 -64.048 -20.800

c5

-5.913 50.863 51.830 -56.002 -87.663 3.699 10.271

-5.913 39.629 6.635 5.453 -68.503 2.422 -42.905 40.134

c6

-5.222 1.942 -18.637 46.151 -11.769 -59.126 -6.696

-5.222 6.263 50.453 -61.475 -34.170 101.073 -108.291 -0.675

φ0 (rad)

source

0.115 1.480 0.570 3.186 6.076 1.718 2.753

ref 57 ref 57 ref 57 ref 58 ref 56 ref 56 this work this work this work this work this work this work this work

0.115 4.847 3.863 1.332 0.000 3.342 4.615 5.199

ref 57 ref 57 ref 57 ref 58 ref 56 ref 56 this work this work this work this work this work this work this work this work

a The first column identifies the four atoms involved in the torsion. The following eight columns show the parameters used in eq 7, either least-squares optimized or converted from OPLS force field.

The optimal LJ and electrostatic scaling factors are both 0.6 for DNB-leucine. The electrostatic scaling factor is quite different from the optimal value for DNB-phenylglycine. However, keeping in mind the impact of the scaling factors on

backbone structure, and the important differences between the optimized structures of both selectors (see Figure 2), the models should differ in significant ways. The Monte Carlo and sequential energy minimization, starting from 33 000 random

Solvation of Chiral Stationary Phases

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13029

TABLE 4: Improper Torsional Potentials for the DNB-phenylglycine and DNB-leucine Modelsa improper torsion

ωe (rad)

kω (kJ mol-1 rad-2)

27, 25, 23, 17, 13, 11,

28, 26, 24, 18, 14, 12,

32, 27, 15, 22, 11, 13,

25 23 25 15 15 10

DNB-phenylglycine -0.02 0.00 0.24 0.00 0.01 -0.10

459.17 435.00 223. 38 531.25 453.56 208.09

25, 23, 21, 13, 11,

26, 24, 22, 14, 12,

30, 25, 15, 15, 10,

23 21 23 11 13

DNB-leucine -0.02 0.00 0.12 -0.01 -0.14

440.15 427.64 290.31 457.75 117.55

a The first column identifies the four atoms defining the improper torsional angle. The second and third columns define the equilibrium angle and the corresponding force constant (see eq 6).

structures, correctly identified the DNB-leucine global minimum and the lowest energy local minimum. The energy difference between these two structures is predicted to be roughly 6 kJ/ mol, a difference that is small but somewhat larger than predicted from ab initio calculations. Importantly, the classical model correctly predicts the presence of additional minima resulting from torsions in the lateral group. In fact, the minimum in Figure 2f results from a twist in the isobutyl group. The classical model correctly identifies this minimum, although it appears within 1 kJ/mol of the global. An additional twisted structure, analogous to the structure found for DNB-phenylglycine, is also predicted for DNB-leucine. With the addition of the tether, and the spatial constraints at the surface, the surface distribution for this structure is expected to be very close to that of the global. II.2. The Interface. Lengthy MD simulations were performed for the DNB-phenylglycine and DNB-leucine surfaces in the presence of four solvents: 100% hexane, 90:10 hexane: 2-propanol, 80:20 hexane:2-propanol, and 100% 2-propanol. In normal-phase chiral chromatography the DNB-phenylglycine and DNB-leucine CSPs are used in conjunction with a hexane: 2-propanol solvent mixture. The percentage of 2-propanol is varied between 0 and 20%, depending on the polarity of the compound that is separated.7,24 We investigate the binary solvent mixtures, but also the 100% hexane and 100% 2-propanol solvents for comparison. To be consistent with experiment, the model interface includes trimethylsilyl end caps and silanol groups along with the selectors. The selectors have a surface coverage of 1.10 µmol/ m2 while the trimethylsilyl end-cap surface density is 3.29 µmol/ m2, consistent with experimental coverage for this type of CSP.22,23 The typical coverage of silanol groups on a silica surface is around 4-8 µmol/m2,23 and a surface density of 4.26 µmol/m2 is chosen for the simulations. The trimethylsilyl end caps correspond directly to a truncated selector tether and their intramolecular potential has the same form as for the tethers: we have chosen OPLS and CHARMM parameters for the Lennard-Jones potential and published potentials for bends and torsions.44,45,57 The OPLS force field46 has been chosen for the silanol groups, and the Rattle algorithm37 is employed to keep the bond lengths constant. The selectors, end caps, and silanol groups are covalently attached to a single underlying layer of Si that is stationary throughout the simulation. This layer is arranged perpendicular to the z-axis of the simulation cell, defines the boundary of the

interfacial system, and is meant to provide a minimal representation of the underlying substrate (see Figure 3a,b). The closest side-by-side distance between two surface molecules (end caps or selectors) is 6.150 Å. II.3. Simulation Details. The simulation cell can be divided into several parts: two parallel surfaces that include selectors, end caps, silanol groups, and an underlying layer of Si; solvent between the surfaces; and empty space above the upper, and below the lower, surface (see Figures 6a and 8a). Consider the solvent. The TraPPE-UA models61 for n-hexane and 2-propanol were chosen. These models were parametrized based on properties of alkane/alcohol mixtures and have been employed in our previous work.18-21 These models are also consistent with eqs 3-5, providing an equivalent level of representation of molecular flexibility throughout the simulation cell. All solvent bond lengths are kept constant using Rattle.37 Ewald summations62 are used to treat the long-ranged Coulombic forces. Following Yeh and Berkowitz,63 a correction for the elongated shape of the simulation cell has been included. The Ewald convergence parameter of R ) 1.4783 nm-1 was chosen for all simulations, with a reciprocal space cutoff of k2 * e 27. The full simulation cell has dimension of 49.92 Å × 49.92 Å × 150 Å, with 72.1-83.6% of this volume as empty space beyond the slabs (depending on the solvent). The distance between the two surfaces is determined by the solvent density.64 Specifically, in order to achieve the correct solvent density in the simulation cell, we account roughly for the volume of the selectors, end caps, and silanol groups in setting the distance between the Si surfaces. Following this, the bulk density in the center of the simulation cell is examined during the simulation to ensure that it is within 5% of the experimental density. Thus, because of inexact estimates of surface volume, the solvent density in the simulation cell will differ from the experimental density. Two Nose´-Hoover thermostats65,66 are used to generate canonical (NVT) properties at 298 K. Independent translational and rotational thermostats are employed, and the conserved quantity is written as

HNH ) Efull +

Qt 2 Qr ζt + gtkT ln st + ζr2 + grkT ln sr (8) 2 2

where Efull is given in eq 1 and the quantities marked with “t” and “r” are for translation and rotation, respectively. The two additional variables, st for translation and sr for rotation, have masses of Qt and Qr,67 respectively, and gt and gr are the corresponding degrees of freedom. The equations of motion are integrated following the algorithm of Martyna et al.68 The time step in each simulation is 0.3 fs, (0.25 for 100% 2-propanol simulations) which is sufficient to limit the change in HNH to less than 1% during the collection period of the simulation. For all four solvents, five 3 500 000 time step molecular dynamics simulations have been performed for both DNBphenylglycine and DNB-leucine interfaces. Our method of generating the starting configuration has been discussed in detail elsewhere.18 Briefly, solvent molecules are randomly placed between the chiral surfaces and strong overlaps are relaxed with Monte Carlo cycles. Once the starting configuration is obtained, the fluid equilibrates for the first 50 000 time steps, leaving a 3 450 000 step collection period. The equilibrium properties reported in section III are averages over 10 surfaces (two surfaces per simulation and five simulations per solvent), reflecting a total collection period of 10.52 ns. All simulations are performed with the MDMC program.69

13030 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Nita and Cann cell, and Lz is the distance between the two surfaces. nact(z) is the number of atoms found between z and z + dz above the underlying Si, while nideal is the number expected for an unstructured fluid. Surface distributions are employed for the selectors, to identify their major conformations at the interface. These distributions are also evaluated for the solvents, where they identify preferred solvent locations and solvent partitioning at the interface. Since the simulation cell includes two surfaces, the surface distributions reported below are averages over both surfaces. Radial distributions, g(r), have been evaluated between potential selector-solvent hydrogen-bonding pairs. These distributions offer a first indication of the importance of hydrogen bonding at specific selector sites. We explore hydrogen bonding in two other ways. First, snapshot analysis coupled with the application of a geometric criterion70,71 of a hydrogen bond provides H-bonding statistics. With regard to the criterion, a hydrogen bond is identified as follows: the distance between H and the hydrogen bond acceptor should be less than 2.6 Å and the angle formed between donor-H-acceptor should be larger than 150°. We report statistics based on an analysis of solvent near 112 000 (5 simulations/solvent × 700 snapshots/simulation × 32 selectors/snapshot) selectors. Two-dimensional (2D) cylindrical distributions g(rc,zc) are also used to study the relative positions of the solvent atoms around the selector atoms. In a 2D distribution the distance between two chosen atoms is divided into two components, rc and zc, according to r ) [(xc2 + yc2) + zc2]1/2 ) (rc2 + zc2)1/2. The variable rc defines the distance between the atoms in a direction parallel to the underlying Si layer, while zc provides the corresponding information perpendicular to the surface. Within the simulation, the 2D distribution is calculated from

g(rc, zc) )

Figure 4. Surface distribution of the lateral groups. Snapshots illustrate the range of conformers for DNB-phenylglycine (a) and DNB-leucine (b) in 90:10 n-hexane:2-propanol. Silicon, oxygen, nitrogen, carbon, and hydrogen are shown in gold, red, blue, gray, and white, respectively. The phenyl ring and isobutyl chain are emphasized in green. The phenyl ring distribution in DNB-phenylglycine is shown with C(17), C(19), and C(21) distributions (filled circles, filled triangles, and open triangles, respectively) in 100% hexane (c), 90:10 n-hexane:2-propanol (e), 80: 20 n-hexane:2-propanol (g), and 100% 2-propanol (i). Similarly, the distribution of the isobutyl chain in DNB-leucine is shown with C(17), C(19), and C(20) distributions (filled circles, filled triangles, and open triangles, respectively) in 100% hexane (d), 90:10 n-hexane:2-propanol (f), 80:20 n-hexane:2-propanol (h), and 100% 2-propanol (j).

II.4. Distribution Functions. Surface distributions provide probabilities for finding atoms at a specified distance above the underlying silicon layer. Within the simulations, these distributions are evaluated from act nact(z) n (z)Lz g(z) ) ideal ) N∆z n

(9)

where z ) 0 corresponds to the underlying immovable, layer of Si atoms, N is the total number of atoms in the simulation

nact(rc, zc) nideal(rc)

)

nact(rc, zc) 2πrc∆rc∆zcF

(10)

where F is the number density of the solvent. These distributions are evaluated every 20 iterations for potential solvent-selector hydrogen-bonding pairs. A positive value for zc indicates that the solvent atom is further from the Si layer than is the selector atom. In other words, the solvent atom is closer to the bulk fluid in the center of the simulation cell. This analysis is particularly instructive for hydrogen bonding as it provides direct information on the solvent position relative to the selector. We will examine 2D distributions, and provide corresponding snapshots, to identify the most probable positions for hydrogen bond donors/acceptors about the selectors. III. Results III.1. Surface Structure. The DNB-phenylglycine and DNBleucine chiral selectors are flexible. As shown in Figure 2, for each selector the global energy minimum and the lowest-energy local minimum are close in energy. This is particularly true for DNB-leucine, where the energy difference is within the error of the B3LYP functional used to find the optimized structures. Each simulation begins with all selectors at the interface given the structure corresponding to the global energy minimum. However, the selectors quickly adopt a range of conformations, and this is evident from Figures 4a,b and 5a,b. These snapshots show the chiral surface, after 1.05 ns, including the trimethylsilyl end caps and the silanol groups. It is worth noting that the end caps and silanol groups are found within 5 Å of the underlying silicon layer. Similarly, the tether atoms are very well localized and found within 7 Å of the Si layer. The solvent is omitted from the snapshot to highlight the selector conformations.

Solvation of Chiral Stationary Phases

Figure 5. Surface distribution of the DNP rings. Snapshots show the range of conformers explored for DNB-phenylglycine (a) and DNBleucine (b) in 90:10 n-hexane:2-propanol. Silicon, oxygen, nitrogen, carbon, and hydrogen are shown in gold, red, blue, gray, and white, respectively. The DNP rings are emphasized in green. The DNP distribution in DNB-phenylglycine is shown with C(27), N(33), and N(36) atoms (filled circles, filled triangles, and open triangles, respectively) in 100% hexane (c), 90:10 n-hexane:2-propanol (e), 80: 20 n-hexane:2-propanol (g), and 100% 2-propanol (i). Similarly, the DNP distribution in DNB-leucine is shown with C(25), N(31), and N(34) atoms (filled circles, filled triangles, and open triangles, respectively) in 100% hexane (d), 90:10 n-hexane:2-propanol (f), 80: 20 n-hexane:2-propanol (h), and 100% 2-propanol (j).

The snapshots in Figures 4 and 5 correspond to a 90:10 n-hexane:2-propanol solvent mixture. Figure 5a shows that the DNB-phenylglycine prefers to place the dinitrophenyl ring roughly parallel to the surface for this solvent composition. This is a bent configuration, similar to the global energy minimum in Figure 2a when the alkyl tether is included. The DNB-leucine

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13031

Figure 6. Simulation results for DNB-phenylglycine in 90:10 n-hexane: 2-propanol. A side view of the simulation cell is presented in (a). Silicon, oxygen, carbon, and hydrogen atoms are shown in gold, red, gray, and white, respectively. Selector molecules are shown in blue and 2-propanol molecules are bolded. (b) Radial distributions between potential selector-solvent H-bonding pairs: H(12)-O(2-propanol), O(14)-H(2-propanol), H(24)-O(2-propanol), and O(26)-H(2-propanol) are represented by filled circles, open squares, filled diamonds, and open triangles, respectively. (c) n-Hexane distribution above the surface: CH3, 2-CH2, and 3-CH2 are represented by filled circles, open squares, and filled diamonds, respectively. (d) Distribution of 2-propanol above the surface: CH3, CH, O, and H are identified by filled circles, filled squares, open diamonds, and open triangles, respectively. The surface distribution of the lower amide atoms is shown in (e). Circles, squares, diamonds, and triangles correspond to N(23), H(24), C(25), and O(26), respectively. The surface distribution of the upper amide atoms is shown in (f). Circles, squares, diamonds, and triangles correspond to N(11), H(12), C(13), and O(14), respectively.

interface in Figure 5b is distinct from the DNB-phenylglycine interface. In particular, the dinitrophenyl rings are often perpendicular to the surface. The lateral groups shown in Figure 4 also adopt preferred positions relative to the underlying surface. Figure 4a shows that the phenyl rings are roughly perpendicular to the surface. The isobutyl group of DNB-leucine prefers to tip toward the surface (Figure 4b). The impact of solvent composition on the surface distributions of the lateral groups and the DNP rings is shown in Figures 4 and 5, respectively. Consider DNB-phenylglycine first. In pure hexane, the solvent does not interact strongly with the selectors and the dinitrophenyl groups are often parallel to the surface. This is evident from Figure 5c, where the distributions show that C(27) and the nitrogens are all most likely to be found at roughly 8 Å above the surface. The lateral phenyl rings, in this solvent, prefer to point toward the bulk, as shown in Figure 4c,

13032 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Nita and Cann

Figure 7. Analysis of hydrogen bonding for the DNB-phenylglycine interface in 90:10 n-hexane:2-propanol. Two-dimensional distribution functions, g(rc,zc), and representative snapshots are shown. The distribution of O(2-propanol) about H(12), H(2-propanol) about O(14), O(2-propanol) about H(24), and H(2-propanol) about O(26) are shown in (a), (b), (c), and (d), respectively. Illustrative snapshots, showing 2-propanol molecules around a single selector, are shown in (e), (f), (g), and (h) with interatomic separations given in angstroms. Silicon, oxygen, nitrogen, carbon, and hydrogen atoms are shown in gold, red, blue, gray, and white, respectively. The 2-propanol molecules are emphasized in the snapshots, and their corresponding positions are shown in panels (a)-(d).

where C(19) and C(21) tend to be further from the surface than C(17). With the addition of 2-propanol, the dinitrophenyl groups are more likely to be found perpendicular to the surface. Specifically, some selectors place their dinitrophenyl groups such that one nitro points toward the surface while the other is directed into the bulk. In a 100% 2-propanol solvent, the latter becomes the dominant conformation for DNB-phenylglycine selectors at the interface. This is evident from Figure 5i, where one nitrogen is closest to the surface, followed by C(27), and then the other nitrogen. The lateral group adopts a similar arrangement relative to the surface: Figure 4i shows that C(21) is closer to the surface than C(17) or C(19). The solvent dependence is equally striking, but quite different, for DNB-leucine. This selector, in pure n-hexane, displays a variety of conformations. The broad range of distances over which the DNP atoms may be found above the surface (Figure 5d) reflects this. The lateral isobutyl group also displays a range of conformations, but Figure 4d indicates that the end carbons (C(19) and C(20) in Figure 1b) tend to be slightly closer to the surface than C(17). The low energy of the local minima found for this selector (see Figure 2) is consistent with these observations: many conformations are energetically accessible. The addition of small quantities of 2-propanol quickly shifts the conformations toward an extended structure where the dinitrophenyl groups point into the bulk. In fact, dinitrophenyl groups are virtually never parallel to the surface for the DNB-leucine interface if the solvent is pure 2-propanol. This is clear from Figure 5j, where both nitrogens of the DNP ring are found further from the surface than C(25). The lateral group also adopts a different configuration with the most probable isobutyl

arrangement corresponding to one end carbon closer to the surface than the other (Figure 4j). Interselector interactions at the interface are rare, despite the potential for hydrogen bonding and ring-ring interactions. This was verified by an analysis of all snapshots. Virtually all interactions occur with negligible probability except perhaps for hydrogen-bonding interactions between amide H and nitro Oʼs of neighboring selectors. In pure n-hexane, roughly 1% of the selectors participate in these hydrogen bonds. However, with the addition of 2-propanol, hydrogen bonds to the solvent replace interselector hydrogen bonding and the probability for these interactions drops below 1%. Ring-ring stacking between neighboring selectors occurred for less than 0.4% of the selectors: a rare occurrence. Intraselector hydrogen bonding was also examined in the snapshots. The potential for hydrogen bonding between the two amide groups in individual selectors was discussed in section II.1. From the snapshots, intramolecular amide-amide H-bonds were not found. Panels e and f in Figures 6 and 8 show the distribution of amide groups above the underlying silicon layer, for DNBphenylglycine and DNB-leucine, respectively. The amide group nearest the surface, atoms 11-14 in Figure 1, is referred to as the “lower” amide group; the other amide linkage is termed the “upper” amide group. The amide atom distributions above the surface are similar for DNB-phenylglycine and DNB-leucine, although the upper group explores a broader range of separations for the latter. For instance, the upper amide oxygen can be found between 6.0 and 12.5 Å above the underlying silicon for DNBleucine, but this atom is only found between 8.0 and 12.5 Å for the DNB-phenylglycine CSP. The peak heights are consis-

Solvation of Chiral Stationary Phases

Figure 8. Simulation results for the DNB-leucine interface in 90:10 n-hexane:2-propanol. A side view of the simulation cell is presented in (a). Silicon, oxygen, carbon, and hydrogen atoms are shown in gold, red, gray, and white, respectively. Selector molecules are shown in blue and 2-propanol molecules are bolded. (b) Radial distributions between potential selector-solvent hydrogen-bonding pairs: H(12)-O(2-propanol), O(14)-H(2-propanol), H(22)-O(2-propanol), and O(24)-H(2propanol) are represented by filled circles, open squares, filled diamonds, and open triangles, respectively. (c) Distribution of n-hexane above the surface: CH3, 2-CH2, and 3-CH2 are represented by filled circles, open squares, and filled diamonds, respectively. (d) Distribution of 2-propanol above the surface: CH3, CH, O, and H are identified by filled circles, filled squares, open diamonds, and open triangles, respectively. The surface distribution of the lower amide atoms is shown in (e). Circles, squares, diamonds, and triangles correspond to N(21), H(22), C(23), and O(24), respectively. The surface distribution of the upper amide atoms is shown in (f). Circles, squares, diamonds, and triangles correspond to N(11), H(12), C(13), and O(14), respectively.

tently lower for DNB-leucine, which also indicates a greater flexibility and delocalization for this selector. Despite the evident structural similarity between DNBphenylglycine and DNB-leucine, the replacement of a lateral phenyl ring by an isobutyl group leads to important structural changes. First, ab initio results identify an interchange between global and lowest-energy local minima. Furthermore, the flexible isobutyl group introduces additional energetically accessible local minima. Within the simulations, both selectors exhibit unique conformational preferences at the interface and the atoms of DNB-phenylglycine are more localized relative to the underlying surface. As discussed above, these selectors respond differently to increasing concentrations of hydrogen-bonding solvent: the alcohol encourages DNB-phenylglycine to orient the dinitrophenyl ring perpendicular to the surface, with one

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13033 nitro group pointing toward the bulk and the other toward the surface, while the DNB-leucine ring points toward the bulk in 100% 2-propanol. III.2. Solvation of DNB-phenylglycine CSP. Figure 6a shows a side-view snapshot from the simulation of the DNBphenylglycine interface with 90:10 n-hexane:2-propanol. The alcohol clearly prefers the interface over the bulk. However, when the alcohol is in the bulk it typically clusters with other alcohols, with hydrogen bonds between them. The alcohols do not descend below the end caps, and this is evident from the snapshot. However, the selectors are well surrounded by 2-propanol. The distribution of alcohol at the interface is more evident from the surface distribution in Figure 6d. The figure shows two peaks beginning at roughly 5 Å, which is directly above the end caps, and ending at roughly 15 Å, above the amide groups and generally above the dinitrophenyl groups. A region of decreased probability occurs between 9 and 10 Å above the underlying surface. The snapshot in Figure 6a identifies the reason for this: Most of the selector atoms are found at, or near, these separations. There is consequently little space available for solvent, and the surface distribution of 2-propanol reflects this crowding. Figure 6c presents the distribution of n-hexane above the DNB-phenylglycine interface. The alcohol prefers the bulk; however, there is a region of enhanced probability around 7 Å from the underlying silicon. The topmost atom of the tether is found at this separation, and the upper amide (Figure 6f), the lateral phenyl ring (Figure 4e), and the dinitrophenyl ring (Figure 5e) are usually found at larger distances from the surface. Thus, between the top of the end caps and the crowded region around 9-10 Å, there is room for solvent. Both 2-propanol and n-hexane are found in this region. The presence of alcohol at the interface strongly suggests that solvent-selector hydrogen bonding is significant for DNBphenylglycine. Radial distributions between potential hydrogenbonding pairs are shown in Figure 6b. The distribution between an amide oxygen (O(14) or O(26)) and the hydrogen of 2-propanol shows a significant region of enhanced probability. Specifically, the probability of finding the alcohol hydrogen within 2 Å of the upper amide oxygen, O(26), is 40 times higher than expected based on the bulk density. The lower amide oxygen, O(14), also has a significant probability of having a hydrogen from 2-propanol nearby. The alcohol can also accept hydrogen bonds from the selector. For the hydrogen in the upper amide, H(24), the probability of finding the alcohol oxygen roughly 2.5 Å away is 7 times higher than expected based on the bulk density. These radial distributions strongly suggest important solvent-selector hydrogen-bonding interactions, especially for the amide oxygens (O(14) and O(26)) of DNBphenylglycine. Further evidence of hydrogen bonding is obtained from the snapshots. Specifically, all 3500 snapshots are examined for the presence of hydrogen bonds between 2-propanol and the DNBphenylglycine selectors. The results of this analysis are summarized in Table 5. Focusing on the results for 90:10 n-hexane: 2-propanol, on average 19.6% of the upper amide oxygen, O(26), are hydrogen bonded while 11.6% of the lower amide oxygens, O(14), have hydrogen bonds. In contrast, 2.6-3.6% of the amide hydrogens have H-bonds on average. The nitro oxygens could hydrogen bond in principle, but snapshot analysis reveals that fewer than 1% of the selectors have H-bonds of this type, on average. This is consistent with the proposed recognition models5,24 for these selectors where the nitro groups do not have a major role in H-bonding, but their role is to make the dinitrophenyl ring a good π-acceptor.

13034 J. Phys. Chem. B, Vol. 112, No. 41, 2008

Nita and Cann

TABLE 5: Hydrogen-Bonding Statistics for DNB-phenylglycine and DNB-leucine Interfacesa percentage percentage percentage occurrence occurrence occurrence in 0:100 in 80:20 in 90:10 hexane: hexane: 2-propanol hexane: selector component component 2-propanol 2-propanol 2-propanol O(14) O(26) H(12) H(24) O(14) and H(24) O(34) or O(35) or O(37) or O(38) O(14) O(24) H(12) H(22) O(14) and H(22) O(24) and H(12) O(32) or O(33) or O(35) or O(36)

DNB-phenylglycine H 11.6 H 19.6 O 2.6 O 3.6 H and O 1.0 H 0.7

11.8 32.7 5.1 4.3 1.1 1.5

18.1 50.5 4.7 3.7 0.9 3.6

DNB-leucine H 21.0 H 24.6 O 27.4 O 33.3 H and O 3.4 H and O 3.7 H 1.4

29.6 35.6 38.6 43.6 4.5 3.8 2.2

54.6 71.8 73.9 49.7 2.5 1.4 5.9

a A structural definition of hydrogen bonding has been applied (see section III), and results are presented for three solvent compositions. The first two columns identify the selector and solvent atoms involved in the hydrogen bonding, and the following columns identify the percentage of the selectors with the specified hydrogen bond.

Both snapshot analysis and radial distributions indicate significant H-bonds to the amide oxygens (O(14) and O(26)), but relatively few H-bonds to the amide hydrogens (H(12) and H(24)). This means that, for a chromatographic separation, the amide hydrogens may be available to the analyte for hydrogen bonding, although steric factors may interfere with this conclusion. The 2D distributions in Figure 7 provide information on the location of the solvent about the hydrogen-bonding groups in the selectors. Consider the 2D distribution of the alcohol oxygen about H(12), the lower amide hydrogen of the selector, shown in Figure 7a. Four distinct regions are apparent: a ridge beginning at zc ) 2 Å and rc ) 0 Å and ending at zc ) 0 Å and rc ) 2 Å; a broad ridge beginning at zc ) 4 Å and rc ) 0 Å; a peak at zc ) 0 Å and rc ) 5.5 Å; and a ridge appearing above rc ) 6 Å. The snapshot in Figure 7e shows solvents distributed around H(12) with zc and rc values given. A 2-propanol directly hydrogen bonding to H(12) is evident in the snapshot and is separated by a distance that corresponds to the ridge beginning at zc ) 2 Å and rc ) 0 Å. Thus, this particular ridge corresponds to direct hydrogen bonding to H(12) and indicates that the H-bonding alcohol is either parallel (zc ) 0 Å) or above (0 Å < zc < 2 Å) the hydrogen, but typically not below. In the snapshot, a second 2-propanol, located at zc ) 4.1 Å and rc ) 1.7 Å, is hydrogen bonding to the upper carbonyl oxygen, O(26). This alcohol contributes to the ridge beginning at zc ) 4 Å and rc ) 0 Å, and from the 2D distribution, it is further from the underlying surface than H(12). A third 2-propanol is shown in the snapshot. This alcohol is hydrogen bonding to the lower carbonyl oxygen, O(14), and its position relative to H(12) indicates that it is the source of the peak at around zc ) 0 Å and rc ) 5.5 Å. The 2D distribution about the lower amide oxygen, O(14), is shown in Figure 7b, and the associated snapshot is provided directly below, in Figure 7f. The alcohol has a high probability to H-bond to O(14), and this is evident from the prominence of the ridge beginning at zc ) 2 Å and rc ) 0 Å in the 2D distribution. This solvent prefers to be further from the underly-

ing surface than O(14), but it may also be slightly closer. The snapshot shows that it may also simultaneously hydrogen bond to another 2-propanol, and this second alcohol contributes to a weak peak in the 2D distribution that appears around zc ) 1 Å and rc ) 4 Å. Finally, an alcohol hydrogen bonded to the upper carbonyl oxygen is evident in the snapshot and contributes to the ridge at zc ) 5 Å. This latter ridge indicates that the alcohol hydrogen, for the 2-propanol hydrogen bonded to the upper carbonyl, is typically 5 Å further from the surface than O(14), but it may be directly above or shifted parallel to the surface. For hydrogen, H(24), a peak at zc ) 0 Å is evident in the 2D distribution of Figure 7c. The presence of a peak, rather than a ridge, in the 2D distribution indicates that the alcohol is fairly localized. In particular, the location of the peak indicates that hydrogen bonding to H(24) occurs when the solvent is on the side, as shown in the snapshot in Figure 7g. It is also possible for a solvent to simultaneously hydrogen bond to H(24) and O(14), and this is evident in the snapshot shown in Figure 7g. A closer inspection of the peak in the 2D distribution reveals that there are, in fact, two peaks contributing in this region. Returning to Table 5, 1% of the selectors have simultaneous H-bonds to a single 2-propanol, via O(14) and H(24). However, on average, only 3.6% of selectors have hydrogen bonds to H(24). Therefore, roughly one quarter of every hydrogenbonding event at H(24) involves a solvent that simultaneously H-bonds to O(14). Both of these events contribute to the peak in the 2D distribution. The 2D distribution of the upper amide oxygen, O(26), shown in Figure 7d has a significant ridge at 0 Å < zc < 2 Å indicating a high probability for hydrogen bonding with the solvent preferring to be above O(26). The other ridge in the 2D distribution, at zc ≈ -4 Å, reflects H-bonding to the lower amide group. The relationship between solvent composition and selector conformations has been discussed previously. The extent and characteristics of solvent-solute hydrogen bonding will also depend on solvent composition. Snapshot analysis has been performed for all solvent compositions, and the results are presented in Table 5. Compare the hydrogen-bonding statistics for 90:10 and 80:20 n-hexane:2-propanol. The addition of more alcohol leads to significantly more hydrogen bonds to the upper amide oxygen, O(26). The frequency of hydrogen-bonding events for other amide atoms also increases, but by a smaller amount, when the solvent includes more alcohol. In a 100% 2-propanol solvent, over 50% of the upper amide oxygens and roughly 18% of the lower amide oxygens have hydrogen bonds to solvent. The hydrogen-bonding probabilities to amide hydrogens are small by comparison and have increased by only a few percent relative to 90:10 n-hexane:2-propanol. Thus the impact of increasing alcohol content in the solvent is different for amide oxygens and amide hydrogens. III.3. Solvation of DNB-leucine CSP. Figure 8 presents a side-view snapshot of the DNB-leucine interface with 90:10 n-hexane:2-propanol. The alcohol clearly prefers the interface, as it also did for DNB-phenylglycine. The extended structure of the DNB-leucine selectors is evident in the snapshot, and relative to DNB-phenylglycine, there is less crowding of the selectors at the interface. Figure 8c,d shows the distribution of solvent as a function of separation from the underlying silicon layer. Both n-hexane and 2-propanol have an increased probability of being found directly above the end caps. However, n-hexane shows a region of decreased probability at around 10 Å while the distribution of 2-propanol is more complex. In particular, the probability

Solvation of Chiral Stationary Phases

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13035

Figure 9. Analysis of hydrogen bonding for the DNB-leucine interface in 90:10 n-hexane:2-propanol. Two-dimensional distribution functions, g(rc,zc), and representative snapshots are shown. The distribution of O(2-propanol) about H(12), H(2-propanol) about O(14), O(2-propanol) about H(22), and H(2-propanol) about O(24) are shown in (a), (b), (c), and (d), respectively. Illustrative snapshots, showing 2-propanol molecules around a single selector, are shown in (e), (f), (g), and (h) with interatomic separations given in angstroms. Silicon, oxygen, nitrogen, carbon, and hydrogen atoms are shown in gold, red, blue, gray, and white, respectively. The 2-propanol molecules are emphasized in the snapshots, and their corresponding positions are shown in panels (a)-(d).

for the alcohol oxygen and hydrogen are still significant at this separation although the CH and CH3 groups are less likely to be found at this distance from the underlying Si. This is in contrast to DNB-phenylglycine, where selector crowding excluded the alcohol from this region. Figure 8b provides radial distributions between potential solvent-selector hydrogen-bonding pairs. As with DNB-phenylglycine, the amide oxygens have a significant probability of finding an alcohol hydrogen roughly 2 Å away. Unlike DNBphenylglycine, both the upper and lower amide oxygens are comparable, with the respective radial distributions showing that approximately 45 times more alcohol oxygen is found at this separation than expected based on the bulk density. Consider the amide hydrogens. For DNB-phenylglycine, solvent Hbonding to these hydrogens occurred infrequently. This is not the case for DNB-leucine, where the radial distributions between these hydrogens and the alcohol oxygen are sharply peaked. Specifically, both the upper and lower hydrogens are 45-55 times more likely to have an alcohol at roughly 2 Å than expected based on the bulk density. Snapshot analysis for hydrogen bonding at the DNB-leucine interface is presented in Table 5. Even in 90:10 n-hexane:2propanol, on average 27-33% of the selectors have a hydrogen bond to one of the amide hydrogens (H(12) and H(22)). In contrast, the amide oxygens (O(14) and O(24)) form hydrogen bonds in 21-25% of the selectors, on average. A comparison with the hydrogen-bonding statistics for DNB-phenylglycine shows that the leucine-based selector forms many more hydrogen bonds with 2-propanol. This increase occurs mostly at the

amide hydrogens, but the hydrogen-bonding probability is also higher for the amide oxygens. Simultaneous hydrogen bonding between a single 2-propanol and multiple atoms in DNB-leucine occurs. The snapshot analysis results in Table 5 indicate that 3.4% of the selectors have simultaneous H-bonds to O(14) and H(22) while 3.7% of the selectors have H-bonds to O(24) and H(12). In both cases, a single solvent molecule forms simultaneous hydrogen bonds to the upper and lower amide groups. Figure 9 shows 2D distributions for hydrogen bonding about DNB-leucine selectors. The distributions for H(12) (Figure 9a), O(14) (Figure 9b), H(22) (Figure 9c), and O(24) (Figure 9d) all show a prominent ridge beginning at rc ) 0 Å and zc ) 2 Å. This ridge indicates the probable location for solvent hydrogen bonding directly to the relevant selector atom. For the lower amide group, the ridge vanishes as zc becomes negative, indicating that the solvent prefers to hydrogen bond from above or on the side. The upper amide groups have a ridge that persists to negative zc, and for O(24), the hydrogen-bonding solvent prefers to be closer to the surface. This is evident in the snapshot in Figure 9h, where the solvent hydrogen bonding to O(24) is closer to the surface. Secondary ridges in the 2D distributions of Figure 9 are also evident. These correspond either to contributions from hydrogen bonding to other groups in the selector, or to solvent-solvent hydrogen bonding where one of the solvents also bonds to a selector atom. Consider the impact of solvent composition on hydrogen bonding at the DNB-leucine interface. The lower half of Table 5 provides the relevant H-bond statistics. As the alcohol content

13036 J. Phys. Chem. B, Vol. 112, No. 41, 2008 of the solvent increases from 90:10 to 80:20 n-hexane:2propanol, all the nonsimultaneous hydrogen-bonding probabilities to amide atoms increase by 9-11%. This is in contrast to DNB-phenylglycine, where an increase in hydrogen bonds was observed primarily for the “upper” amide oxygen. When the solvent is 100% 2-propanol, the probability for hydrogen bonding is between 50 and 74% for the amide groups. Thus solvent-selector hydrogen bonding is extensive. Compare the hydrogen bonding of the lower amide hydrogen, H(12), in 100% 2-propanol for DNB-phenylglycine and DNB-leucine. For the former, 4.7% of these hydrogens have H-bonds, whereas 73.9% have H-bonds for DNB-leucine. Clearly, these two selectors have very different hydrogen-bonding characteristics at the interface. IV. Conclusions The solvation of leucine- and phenylglycine-based chiral stationary phases is the focus of this article. These CSPs were first synthesized by Pirkle et al. in the 1980s. Ensuing experimental studies showed that these two CSPs displayed distinct separation characteristics.5-7 The difference between these CSPs is somewhat surprising given that leucine and phenylglycine differ only by the replacement of an isobutyl group by a phenyl. In this article, we begin by exploring the stable conformations for DNB-leucine and DNB-phenylglycine. We find that the phenylglycine-based selector has an energetically accessible local energy minimum 11 kJ/mol higher in energy than the global minimum. For the leucine-based selector, two nearly isoenergetic minima are identified and torsions within the lateral isobutyl group will lead to several other energetically accessible conformations. Thus these compounds have significant differences in their conformations. In order to properly represent the flexibility of these selectors, potentials were carefully derived. First, the energy was evaluated, at the B3LYP/6-311G** level of theory, as the molecule explores a range of conformations. Comparisons were also made with MP2/6-311G** and HF/6-311G** conformers. Second, the intramolecular potential parameters were obtained by leastsquares fitting. In particular, torsions were all fitted simultaneously to the ab initio results. Finally, the incorporation of nonbonding intramolecular interactions has a crucial impact on the conformations predicted from the potential. The scaling parameters that determine the strength of these interactions were systematically varied and optimized by comparison between the corresponding minimum-energy structures and the ab initio structures. Simulations of the chiral interfaces were performed in 100% n-hexane, 90:10 n-hexane:2-propanol, 80:20 n-hexane:2-propanol, and 100% 2-propanol. We find that the DNB-phenylglycine selectors prefer to orient such that the dinitrophenyl groups are either parallel to the surface or oriented such that one nitro group points toward the surface while the other points into the bulk. As the alcohol concentration increases, the latter conformation increases in importance. For DNB-leucine selectors, the dinitrophenyl ring tends to either lie parallel to the surface or point directly into the fluid. The latter becomes more prevalent with an increased presence of 2-propanol. For the mixed solvents, the alcohol strongly prefers the interfacial region and can be found directly above the end caps and around the selectors. For the phenylglycine-based selector, the alcohol is not as likely to be found within 9-10 Å above the underlying Si layer. This occurs because of crowding from the selectors in this separation range. The leucine-based selectors

Nita and Cann TABLE 6: Number of Hydrogen Bonds per Selector in 90:10, 80:20, and 0:100 n-Hexane:2-Propanol Solvents no. of H-bonds per selector molecule

percentage occurrence in 90:10 hexane: 2-propanol

percentage occurrence in 80:20 hexane: 2-propanol

percentage occurrence in 0:100 hexane: 2-propanol

0 1 2 3 4

DNB-phenylglycine 68.0 55.8 27.0 35.9 4.5 7.8 0.4 0.5 0.0 0.0

38.1 48.5 11.9 1.4 0.1

0 1 2 3 4

DNB-leucine 35.2 19.8 33.7 34.1 22.6 29.2 6.8 12.8 1.7 4.1

2.4 14.7 31.9 33.3 17.1

are more extended, and this crowding is less severe. As a result, 2-propanol shows less depletion between 9 and 10 Å. Extensive hydrogen bonding between the 2-propanol and the selectors has been found. For DNB-leucine both the amide oxygens (O(14) and O(24)) and the amide hydrogens (H(12) and H(22)) participate in H-bonding with the solvent. In 100% 2-propanol, each of these four atoms has a 50-74% probability for H-bonding to the alcohol. In contrast, DNB-phenylglycine in 100% 2-propanol has between 3.7 and 50.5% probability for H-bonding, with the amide hydrogens only H-bonding infrequently. In other words, DNB-phenylglycine selectors typically do not have H-bonding between the amide hydrogens and solvent. This is not the case for DNB-leucine, where these H-bonding interactions are frequent. Table 6 summarizes the number of solvent-selector H-bonds observed, on average, for individual selectors. Consider 90:10 n-hexane:2-propanol. For the DNB-phenylglycine interface, most selectors do not have any H-bonds although 27% will have one H-bond, on average, and this bond will typically involve an amide oxygen. For DNBleucine, 33.7% of the selectors have one H-bond, 22.6% have two H-bonds, and 6.8% have three H-bonds. Thus, a much more significant fraction of the selectors has H-bonds for DNB-leucine and all four amide atoms (H(12), O(14), H(22), O(24)) participate in H-bonding. Acknowledgment. The financial support of the Natural Sciences and Engineering Research Council (NSERC) of Canada is gratefully acknowledged. The authors acknowledge the generous allocation of computer resources by the High Performance Computing Virtual Laboratory (HPCVL), Westgrid, and SHARCNET. References and Notes (1) Ward, T. J.; Hamburg, D. M. Anal. Chem. 2004, 76, 4635. (2) Pirkle, W. H.; Welch, C. J. J. Org. Chem. 1984, 49, 138. (3) Welch, C. J. J. Chromatogr. 1994, 666, 3. (4) Pirkle, W. H.; House, D. W.; Finn, J. M. J. Chromatogr. 1980, 192, 143. (5) Pirkle, W. H.; Pochapsky, T. C. Chem. ReV. 1989, 89, 347. (6) Pirkle, W. H.; Finn, J. M.; Schreiner, J. L.; Hamper, B. C. J. Am. Chem. Soc. 1981, 103, 3964. (7) Berthod, A.; Jin, H. L.; Stalcup, A. M.; Armstrong, D. W. Chirality 1990, 2, 38. (8) Regis Technologies Inc.; www.registech.com. (9) Pirkle, W. H.; Welch, C. J.; Zych, A. J. J. Chromatogr. 1993, 648, 101. (10) Pirkle, W. H.; McCune, J. E. J. Chromatogr. 1989, 469, 67. (11) Wainer, I. W.; Doyle, T. D. J. Chromatogr. 1984, 284, 117.

Solvation of Chiral Stationary Phases (12) Hyun, M. H.; Kim, Y. D.; Han, S. C.; Lee, J. B. J. High Resolut.Chromatogr. 1998, 21, 464. (13) Hellriegel, C.; Skogsberg, U.; Albert, K.; Lammerhofer, M.; Maier, N. M.; Linder, W. J. Am. Chem. Soc. 2004, 126, 3809. (14) Katte, S. J.; Beck, T. L. J. Phys. Chem. 1993, 97, 5727. (15) Slusher, J. T.; Mountain, R. D. J. Phys. Chem. B 1999, 103, 1354. (16) Cheng, C. Y.; Chen, T. L.; Wang, B. C. J. Mol. Struct. (THEOCHEM) 2002, 577, 81. (17) Zhang, L.; Sun, L.; Siepmann, J. I.; Schure, M. R. J. Chromatogr., A 2005, 127, 1079. (18) Nita, S.; Cann, N. M.; Horton, J. H. J. Phys. Chem. B 2004, 108, 3512. (19) Nita, S.; Cann, N. M.; Horton, J. H. J. Phys. Chem. B 2006, 110, 9511. (20) Zhao, C.; Cann, N. M. J. Chromatogr., A 2006, 1131, 110. (21) Zhao, C.; Cann, N. M. J. Chromatogr., A 2007, 1149, 197. (22) Szczerba, T. J. Regis Technologies, Inc.; personal communication. (23) Nawrocki, J. J. Chromatogr., A 1997, 779, 29. (24) Pirkle, W. H.; Pochapsky, T. C.; Mahler, G. S.; Field, R. E. J. Chromatogr. 1985, 348, 89. (25) Magora, A.; Abu-Lafi, S.; Levin, S. J. Chromatogr., A 2000, 866, 183. (26) Lipkowitz, K. B. Acc. Chem. Res. 2000, 33, 555. (27) Lipkowitz, K. B. J. Chromatogr., A 2001, 906, 417. (28) Lipkowitz, K. B.; Pearl, G.; Coner, B.; Peterson, M. A. J. Am. Chem. Soc. 1997, 119, 600. (29) Altomare, C.; Carotti, A.; Cellamare, S.; Fanelli, F.; Gasparrini, F.; Villani, C.; Carrupt, P. A.; Testa, B. Chirality 1993, 5, 527. (30) Suzuki, T.; Timofei, S.; Iuoras, B. E.; Uray, G.; Verdino, P.; Fabian, W. M. F. J. Chromatogr., A 2001, 922, 13. (31) Fabian, W. M. F.; Stampfer, W.; Mazur, M.; Uray, G. Chirality 2003, 15, 271. (32) Lipkowitz, K. B. J. Chromatogr., A 1994, 666, 493. (33) Lipkowitz, K. B.; Malik, D. J.; Darden, T. Tetrahedron Lett. 1986, 27, 1759. (34) Lipkowitz, K. B.; Demeter, D. A.; Zegarra, R.; Larter, R.; Darden, T. J. Am. Chem. Soc. 1988, 110, 3446. (35) Lipkowitz, K. B.; Baker, B. Anal. Chem. 1990, 62, 770. (36) Topiol, S.; Sabio, M.; Moroz, J.; Caldwell, W. B. J. Am. Chem. Soc. 1988, 110, 8367. (37) Andersen, H. C. J. Comput. Phys. 1983, 52, 24. (38) Ryckaert, J. P.; Bellemans, A. Faraday Discuss. Chem. Soc. 1978, 66, 95. (39) Lipkowitz, K. B.; Demeter, D. A.; Parish, C. A.; Landwer, J. M. J. Comput. Chem. 1987, 8, 753. (40) Chin, W.; Piuzzi, F.; Dimicoli, I.; Mons, M. Phys. Chem. Chem. Phys. 2006, 8, 1033. (41) Gung, B. W.; MacKay, J. A.; Zou, D. J. Org. Chem. 1999, 64, 700. (42) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Montgomery, J. A., Jr.; Vreven, T.; Kudin, K. N.; Burant, J. C.; Millam, J. M.; Iyengar, S. S.; Tomasi, J.; Barone, V.; Mennucci, B.; Cossi, M.; Scalmani, G.; Rega, N.; Petersson, G. A.; Nakatsuji, H.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; Honda, Y.; Kitao, O.; Nakai, H.; Klene, M.; Li, X.; Knox, J. E.; Hratchian, H. P.; Cross, J. B.; Bakken, V.; Adamo, C.;

J. Phys. Chem. B, Vol. 112, No. 41, 2008 13037 Jaramillo, J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli, C.; Ochterski, J. W.; Ayala, P. Y.; Morokuma, K.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Zakrzewski, V. G.; Dapprich, S.; Daniels, A. D.; Strain, M. C.; Farkas, O.; Malick, D. K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Ortiz, J. V.; Cui, Q.; Baboul, A. G.; Clifford, S.; Cioslowski, J.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz, P.; Komaromi, I.; Martin, R. L.; Fox, D. J.; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Challacombe, M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong, M. W.; Gonzalez, C.; Pople, J. A. Gaussian 03, revision C.02; Gaussian, Inc.: Wallingford, CT, 2004. (43) Still, M. G.; Rogers, L. B. Talanta 1990, 37, 599. (44) Lopes, P. E. M.; Murashov, V.; Tazi, M.; Demchuk, E.; MacKerell, A. D. J. Phys. Chem. B 2006, 110, 2782. (45) Stubbs, J. M.; Potoff, J. J.; Siepmann, J. I. J. Phys. Chem. B 2004, 108, 17596. (46) Jorgensen, W. L.; Tirado-Rives, J. J. Am. Chem. Soc. 1988, 110, 1657. (47) Price, M. L. P.; Ostrovsky, D.; Jorgensen, W. L. J. Comput. Chem. 2001, 22, 1340. (48) Duffy, E. M.; Kowalczyk, P. J.; Jorgensen, W. L. J. Am. Chem. Soc. 1993, 115, 9271. (49) Lorentz, H. A. Ann. Phys. 1981, 12, 127. (50) Berthelot, D. C. Hebd. Se´anc. Acad. Sci., Paris 1988, 126, 1703. (51) Breneman, C. M.; Wiberg, K. B. J. Comput. Chem. 1990, 11, 361. (52) Kony, D.; Damm, W.; Stoll, S.; Van Gunsteren, W. F. J. Comput. Chem. 2002, 23, 1416. (53) De Oliveira, O. V.; Freitas, L. C. J. Mol. Struct. 2005, 728, 179. (54) E04USF, The NAG Fortran Library, Mark 21; Numerical Algorithms Group: Oxford, U.K., 2005. (55) Berardi, R.; Carnelli, G.; Galletti, P.; Giacomini, D.; Gualandi, A.; Muccidi, L.; Zannoni, C. J. Am. Chem. Soc. 2005, 127, 699. (56) Jorgensen, W. L.; Madura, J. D.; Swenson, C. J. J. Am. Chem. Soc. 1984, 106, 6638. (57) Briggs, J. M.; Jorgensen, W. L. J. Comput. Chem. 1990, 11, 958. (58) Jorgenson, W. L. J. Phys. Chem. 1986, 90, 1276. (59) Levine, I. N. Quantum Chemistry, 5th ed.; Prentice-Hall Inc., 2000. (60) Gallagher, J. F.; Kenny, P. T. M.; O’Donohoe, M. Acta Crystallogr. 2000, C56, e157. (61) Chen, B.; Potoff, J. J.; Siepmann, J. I. J. Phys. Chem. B 2001, 105, 3093. (62) Ewald, P. Ann. Phys. (Leipzig) 1921, 64, 253. (63) Yeh, I. C.; Berkowitz, M. L. J. Chem. Phys. 1990, 11, 3155. (64) Morrone, S. R.; Francesconi, A. Z. J. Chem. Thermodyn. 1996, 28, 935. (65) Nose´, S. J. Chem. Phys. 1984, 81, 511. (66) Hoover, W. G. Phys. ReV. A 1985, 31, 1695. (67) Frenkel, D.; Smit, B. Understanding Molecular Simulations: From Algorihms to Applications; Academic: San Diego, CA, 1996. (68) Martyna, G. J.; Tobias, T. J.; Klein, M. L. J. Chem. Phys. 1994, 101, 4177. (69) Cressman, E.; Das, B.; Dunford, J.; Ghenea, R.; Huh, Y.; Nita, S.; Paci, I.; Wang, S.; Zhao, C.; Cann, N. M. Unpublished. (70) Levitt, M.; Sharon, R. Proc. Natl. Acad. Sci. U.S.A. 1988, 85, 7557. (71) Luzar, A.; Chandler, D. Phys. ReV. Lett. 1996, 76, 928.

JP710392K